Dynamics of complex unicritical polynomials
A Dissertation Presented
by
Davoud Cheraghi
to
The Graduate School
in Partial Fulfillment of the
Requirements
for the Degree of
Doctor of Philosophy
in
Mathematics
Stony Brook University
August 2009 Stony Brook University
The Graduate School
Davoud Cheraghi
We, the dissertation committee for the above candidate for the Doctor of
Philosophy degree, hereby recommend acceptance of this dissertation.
Mikhail M. Lyubich Professor of Mathematics, Stony Brook University Dissertation Advisor
John Milnor Distinguished Professor of Mathematics, Stony Brook University Chairperson of Defense
Dennis Sullivan Distinguished Professor of Mathematics, Stony Brook University Inside Member
Saeed Zakeri Assitant Professor of Mathematics, City University of New York Outside Member
This dissertation is accepted by the Graduate School.
Lawrence Martin Dean of the Graduate School
ii Abstract of the Dissertation
Dynamics of complex unicritical polynomials
by
Davoud Cheraghi
Doctor of Philosophy
in
Mathematics
Stony Brook University
2009
In this dissertation we study two relatively independent problems in one dimensional complex dynamics. One on the parameter space of unicritical polynomials and the other one on measurable dynamics of certain quadratic polynomials with positive area Julia sets. It has been conjectured that combinatorially equivalent non-hyperbolic uni- critical polynomials are conformally conjugate. The conjecture has been al- ready established for finitely renormalizable unicritical polynomials. In the first part of this work we prove that a type of compactness, called a priori bounds, on the renormalization levels of a unicritical polynomial, under a cer- tain combinatorial condition, implies this conjecture. It has been recently shown that there are quadratic polynomials with pos- itive area Julia set. The second part of this work investigates typical trajecto- ries of quadratic polynomials with a non-linearizable fixed point of high return
iii type. This class contains the examples with positive area Julia set mentioned above. In particular, we prove that Lebesgue almost every point in the Julia set of these maps accumulates on the non-linearizable fixed point. We show that the post-critical sets of these maps, which are the measure theoretic at- tractors, have measure zero and are connected subsets of the plane. This has some interesting corollaries such as, almost every point in the Julia set of such maps is non-recurrent, or, there is no finite absolutely continuous invariant measure supported on the Julia set of these maps.
iv Contents
ListofFigures...... vii Acknowledgements ...... viii
1 Introduction 1
2 Preliminaries 11 2.1 JuliaandFatousets ...... 11 2.2 Dynamics of polynomials ...... 14 2.3 Unicritical family and the connectedness locus ...... 15 2.4 Quasi-conformalmappings ...... 20 2.5 Polynomial-like maps ...... 22 2.6 Holomorphicmotions...... 24
3 Combinatorial Rigidity in the Unicritical Family 25 3.1 Modifiedprincipalnest ...... 25 3.1.1 Yoccozpuzzlepieces ...... 25 3.1.2 The complex bounds in the favorite nest and renormal- ization...... 26 3.1.3 Combinatoricsofamap ...... 32 3.2 Proofoftherigiditytheorem...... 35 3.2.1 Reductions ...... 35
v 3.2.2 Thurstonequivalence ...... 37
3.2.3 The domains Ωn,j and the maps hn,j ...... 41
i 3.2.4 The gluing maps gn(k) ...... 58 3.2.5 Isotopy...... 62 3.2.6 Promotiontohybridconjugacy ...... 67 3.2.7 Dynamical description of the combinatorics ...... 69
4 Typical Trajectories of complex quadratic polynomials 72 4.1 Accumulationonthefixedpoint...... 72 4.1.1 Post-criticalsetasanattractor ...... 72 4.1.2 The Inou-Shishikura class and the near-parabolic renor- malization...... 75 4.1.3 Sectorsaroundthepost-criticalset ...... 80 4.1.4 Sizeofthesectors...... 89 4.1.5 PerturbedFatoucoordinate ...... 93 4.1.6 Proof of main technical lemmas ...... 106 4.2 Measureandtopologyoftheattractor ...... 111 4.2.1 Balls in the complement ...... 111 4.2.2 Going down the renormalization tower ...... 116 4.2.3 Going up the renormalization tower ...... 121 4.2.4 Proofofmainresults ...... 132 Bibliography ...... 135
vi List of Figures
2.1 TheMandelbrotset...... 17 2.2 The connectedness locus and a primary limb in it ...... 18 M3 2.3 A primitive renormalizable Julia set and the corresponding little multibrotcopy ...... 19
3.1 The multiply connected domains and the buffers ...... 41 3.2 Primitivecase...... 43 3.3 An infinitely renormalizable Julia set ...... 50 3.4 A twice satellite renormalizable Julia set ...... 53
4.1 A schematic presentation of the Polynomial P ...... 76 4.2 An example of a perturbed Fatou coordinate and its domain. . . 78 4.3 Thefirstgenerationofsectors ...... 82
4.4 The domain ΣC2 ...... 100
vii Acknowledgements
First of all, I would like to thank my adviser Mikhail Lyubich for his great patience and support during my graduate studies. He introduced many inter- esting subjects in holomorphic dynamics to me and was particularly helpful with improving my mathematics writing. Especially, his clear lectures have al- ways been a source of enthusiasm for me and for many others in the dynamics community at the Stony Brook university. I would like to thank the people at the Mathematics Department and IMS at Stony Brook for their generous support. I have enjoyed talking to and learning from many friends there. In particular, I want to thank M. Ander- son, C. Bishop, C. Cabrera, H. Hakobian, M. Martens, J. Milnor, S. Sullivan, S. Sutherland, S. Zakeri. R. Roeder was particularly helpful with improving my mathematics and English writing. I had privilege of spending a year of my studies at the Fields Institute in Toronto. I would like to thank them for their support. Specially, many lengthy conversations with C. Chamanara, J. Kahn, R. Perez, D. Schleicher, M. Shishikura, M. Yampolsky at the Institute created excellent memories dur- ing that year. Many thanks go to my great family to whom I owe every thing I have accomplished. Especially, to my parents for their trust in me and the freedom they gave me during all my life, to my brother and sisters for their constant encouragement and the confidence they gave me. Chapter 1
Introduction
In this work we study two different problems in one dimensional holomorphic dynamics. The first one is on combinatorial rigidity of a class of unicritical maps. The second one describes typical (with respect to the Lebesgue measure) trajectories of certain quadratic polynomials with positive area Julia set. Below we describe the content of these two studies. Part 1. Dynamics of a rational map on the Riemann sphere breaks into two pieces; dynamics on the Fatou set and dynamics on the Julia set. Fatou set of a map is “nicely behaved” part of the dynamics, and it is completely understood through works of Fatou [Fat19], [Fat20a], [Fat20b], Julia [Jul18] and Sullivan [Su85]. The Julia set of a rational map is the “chaotic” part of the dynamics. Substantial work by many people has been devoted to under- standing these two sets for different families of rational maps. Basic properties of these sets has been collected in several books and surveys: See for example Beardon [Be91], Milnor [M06], Blanchard [Bl84], and Lyubich [Ly86]. Hyperbolic maps form a class of thoroughly understood and well behaved maps. These are rational maps for which the orbits of all critical points tend to attracting periodic points. For every map in this class, there exists a finite
1 subset of the Riemann sphere such that a full measure set of points on the sphere is attracted to this set under dynamics of the map. One of the main conjectures in holomorphic dynamics, which goes back to Fatou, states that the hyperbolic maps are dense in the space of rational maps of degree d (also in the space of polynomials of degree d). By an approach developed through the work of R. M˜ane, P. Sad and D. Sullivan [MSS83] for rational maps to tackle this problem, “studying families of rational maps has been reduced to studying of dynamics of individual maps on their Julia set”. Every polynomial with one critical point, which is referred as unicritical polynomial through this thesis, can be written (up to conformal conjugacy) in the form z zd + c, for some complex number c. If the critical point → 0 tends to infinity under iteration of such a map (attracted to the super- attracting fixed point at infinity), the map is hyperbolic. The Julia set is a Cantor set in this case. Therefore, attention reduces to parameters for which orbit of 0 stays bounded under iteration. The set of such parameters is called the connectedness locus or the Mulribrot set, (the well-known Mandelbrot set corresponds to d = 2) see Figures 2.1 and 2.2. There is a way of defining graded partitions of the Multibrot set into pieces such that dynamics of maps in each piece have some special combinatorial property. All maps in a given piece of partition of a certain level will be called combinatorially equivalent up to that level. Conjecturally, combinatorially equivalent (up to all levels) non-hyperbolic maps in the uniciritcal family are conformally conjugate. As stated in [DH82] for d = 2, this Rigidity Conjecture is equivalent to the local connectivity of the Mandelbrot set (MLC for short) and naturally extends to degree d unicritical polynomials. In [DH82], they prove that MLC implies Density of hyperbolic quadratic polynomials among
2 quadratic polynomials. These discussions have been extended to degree d unicritical polynomials by D. Schleicher in [Sch04]. Roughly speaking, the rigidity conjecture says that combinatorics of a map in this class uniquely determines (fine scale) geometry of the chaotic part of the dynamics of that map. In particular, combinatorics of such map determine the parameter c up to a finite number of values. A unicritical polynomial f is called Douady-Hubbard renormalizable, or renormalizable for short, if there exist a neighborhood U of the critical point
n and an integer n > 1, such that f ◦ : U V is a degree d proper branched → nk covering with U compactly contained in V , and f ◦ (0) U, for k =0, 1, 2,... . ∈ n 1 By [DH85], such a map f ◦ : U V , denoted by (f), is conjugate to a uni- → R critical polynomial, which we denote it by 1(f), of the same degree as the SR one of f. Moreover, this renormalization, under a certain combinatorial condi- tion which will be clear later, provides a homeomorphism from a subset of the connectedness locus onto all of the connectedness locus. Such homeomorphim is determined up finite number of rotations, however, one can make it unique by marking a particular point on the connectedness locus (which will be made clear later). We will denote the corresponding maximal homeomorphic copy of the connectedness locus (domain of this homeomorphism) within the actual connectedness locus by τ(f). By maximal homeomorphic copy we mean that it is not strictly contained in any other homeomorphic copy except the actual connectedness locus. Given a renormalizable unicritical polynomial f, if (f) is also renormal- R izable we say that f is twice renormalizable. So, a unicritical polynomial is called infinitely renormalizable if the renormalization process defined above can be carried out infinite number of times, that is, the sequence (f), ( (f)), R R R
3 ( ( (f))),... is well defined. Combinatorics of an infinitely renormalizable R R R unicritical polynomial is defined as the sequence of maximal homeomorphic copies τ(f), τ( 1(f)), τ(( )2(f)),... of the connectedness locus. SR SR In 1990’s, J. C. Yoccoz [H93] proved MLC conjecture at all non-hyperbolic parameter values which are at most finitely renormalizable. He also proved lo- cal connectivity of the Julia set for such maps with all periodic points repelling. Degree 2 assumption was essential in his argument. Local connectivity of Julia sets for unicritical degree d polynomials which are at most finitely renormal- izable, and with all their periodic points repelling has been established by J. Kahn and M. Lyubich [KL05] in 2005. Their proof is based on “controlling” geometry of a Modified principal nest. The same controlling technique is used to settle the rigidity problem for these parameters in [AKLS05]. The rigidity conjecture for quadratic maps z2 +c with c real is proved inde- pendently by M. Lyubich [Ly97] and J. Graczyk, G. Sw¸aetek´ [GS98]. Recently, the conjecture has been proved for polynomials with all their critical points real and non-degenerate by O. Kozlovski, W. Shen and S. Van Strien in [KSvS07]. An infinitely renormalizable unicritical polynomial satisfies a priori bounds condition, a notion introduced by D. Sullivan [S92], if infimum of moduli of
annuli V U is non-zero, where ( n(f))tn : U V is a renormalization of n \ n R n → n n(f). Here we prove that this property, under certain combinatorial assump- R tion, implies the rigidity conjecture. We say that an infinitely renormalizable map satisfies the secondary limbs condition, denoted by for short, if the SL maximal connectedness loci copies τ(f), τ( 1(f)),... in the definition of the R combinatorics of f belong to finite number of secondary limbs of the connect- edness locus.
Theorem 1.1 (Rigidity). If fc is an infinitely renormalizable degree d uni-
4 critical polynomial, with a priori bounds, and satisfies condition, then it is SL combinatorially rigid.
This result was proved in part II of [Ly97] for degree 2 polynomials. The main difference between our proof and the one in [Ly97] is in the construction of the Thurston conjugacy. In [Ly97] such a conjugacy is built along the whole principal nest and uses linear growth of moduli along this nest. However, such a growth is not known for arbitrary degree unicritical polynomials. Certain annuli in the modified principal nest introduced in [KL05] have definite module and this helps us to “pass” over principal nest much easier. This makes the whole construction simpler and more general to include arbitrary degree d. To prove this statement, we first construct a quasi-conformal map of the plane which conjugates the two combinatorially equivalent maps on their post- critical sets. It does not respect the dynamics outside of the post-critical sets, but it is in a “right” homotopy class of homeomorphisms of the complex plane minus the post-critical sets. Verifying such a topological property with the rich possibilities of combinatorics of these maps is the source of difficulty in the proof. Being in such a homotopy class enables one to lift this quasi-conformal map infinite number of times (via the two polynomials) and obtain an actual quasi-conformal conjugacy in the limit. Finally, this quasi-conformal conjugacy can be promoted to a conformal conjugacy by an open-closed argument. Different classes of maps are known to enjoy a priori bounds property; real infinitely renormalizable unicritical polynomials (see [LvS98] and [LY97]), primitive infinitely renormalizable maps of bounded type in [K06], parameters under a Decorations condition in [KL06], and in [KL07] under a Molecule condition. The class of maps satisfying the molecule condition contains the primitive infinitely renormalizable parameters of bounded type and the ones
5 satisfying decoration condition. We will see that the combinatorial class SL contains the combinatorial class of all these parameters. Therefore, combining the above theorem with [KL07] we obtain the following:
Corollary 1.2. Infinitely renormalizable, combinatorially equivalent, quadratic polynomials satisfying the molecule condition are conformally equivalent.
Part 2. Let f : Cˆ Cˆ be a rational map of the Riemann sphere. A → central problem in dynamics of a system is to understand behaviour of orbit of a typical point; z, f(z), f 2(z),...,f n(z),... . Indeed, this is a rather old sub- ject in holomorphic dynamics starting with contributions of Koenigs [Ko84], Schr¨oder [Sch71], B¨ottcher [B¨o04] in the late 19th century on the local dynam- ics of holomorphic maps. Let f : U C C be a holomorphic map defined ⊆ → on some neighborhood of z0 with f(z0) = z0 and multiplier f ′(z0) = λ. If λ = 0, 1, by [Ko84], it is known that there exists a local conformal change | | 1 of coordinate near z , with ϕ(z ) = 0, and ϕ f ϕ− (z) = z λz near 0 0 ◦ ◦ → 2π p i zero. There is a similar normal form by [Le97] and [B¨o04] when λ = e q , and λ = 0, respectively. See [M06] for these results and more. This describes orbits near such fixed points. If the multiplier at a fixed point is λ = e2παi with α R Q, f is called lin- ∈ \ earizable at that fixed point, if there exists a conformal change of coordinate on a neighborhood of the fixed point so that in the new system of coordinate the map becomes the linear map z e2παiz. If there is such a domain of lineariza- → tion, the maximal domain of linearizability is called Siegel disk, otherwise the fixed point is called Cremer fixed point. G. Pfeiffer [Pf17] found first examples of rational maps with a non linearizable fixed point, and H. Cremer [Cre38] proved that for a Baire generic choice of α in (0, 1) every rational map with a fixed point of multiplier e2παi is non linearizable. C. L. Siegel [Sie42] showed 6 that for Lebesgue almost every α in (0, 1) every germ with a fixed point of mul- tiplier e2παi is linearizable. However, these results did not cover all rotations. It was apparent from Cremer and Siegel’s proofs that the answer to this problem depends on the arithmetic nature of α. Let
1 α = [a , a , a ,... ] := 1 2 3 a + 1 1 a2+...
pn denote the continued fraction expansion of α, and = [a1, a2, , an] denote qn its rational approximants. Following Siegel’s ideas, Brjuno [Brj71] proved that given approximants pn of α, under the weaker condition qn
∞ log(q ) n+1 < , q ∞ n=1 n every germ with a fixed point of multiplier e2παi is locally linearizable. We say that a real number α is Brjuno, and write α B, if the above series is finite. ∈ Yoccoz [Yoc95] showed that the Brjuno condition is sharp for the quadratic maps. In other words, if α / B, P (z)= e2παiz + z2 is not linearizable at 0. ∈ α Now assume that f : C C is a quadratic polynomial. As the dynamics → on the Fatou set is very well understood, we have a complete understanding of typical orbits of a map, if Julia set of that map has zero area. Indeed, this is known in several cases including
if f is hyperbolic, •
if f has a parabolic cycle, see [DH82] or [Ly83b], •
if f is at most finitely renormalizable with all periodic points repelling, • see [Ly91] or [Sh91],
if f has a Siegel disk with rotation number satisfying log a = (√n), • n O see [PZ04], 7 In particular, siegel disks of bounded type have been studied in greater detail. In [Mc98] and [Ya08] two different types of renormalization technique have been used to describe geometry of the coresponding Julia sets and the dynamics of the map on them. However, X. Buff and A. Ch´eritat [BC05] in 2005, by completing a program initiated by A. Douady, proved that there are quadratic polynomials with a Cremer fixed point and positive area Julia set. This motivates the problem of describing typical orbits of points in the Julia set of such quadratic poly- nomials. In the presence of a Cremer fixed point, there is only one Fatou component, the basin of infinity, which is the set of points that tend to the at- tracting fixed point at infinity under iteration. The Julia set is known to have a complicated topology. For example, it is a non locally connected subset of the plane. Furthermore, by Ma˜n´e[Ma83], the finite critical point is recurrent and its orbit accumulates on the Cremer fixed point.
Given N > 0, let IrrN denote the set of irrational numbers α = [a1, a2,... ], with a N for i =1, 2,... . Our first result in this direction is that i ≥
Theorem 1.3. There exists a constant N such that if α Irr is a non- ∈ N Brjuno number, then orbit of almost every point in the Julia set of Pα(z) = e2παiz + z2 accumulates on the 0 fixed point.
The statement of the above theorem is trivial if the Julia set has zero area. However, the positive area Julia sets constructed by Buff and Ch´eritat fall into class of maps satisfying our conditions in the above theorem. Thus, above theorem applies to these examples. Proof of this theorem uses a Douady-Ghys type renormalization; return map to a certain domain. In a fruitful work by H. Inou and M. Shishikura [IS06] in 2005 this renormalization was developed to study large iterate of maps
8 near fixed points of high return times. They have introduced a compact (in compact-open topology) class of maps with “large enough” domain of definition which is invariant under this renormalization. More precisely, consider P (z)= z(1+z)2 : U C, where U is a certain domain containing the 0 fixed point and → the 1/3 critical point. The Inou-Shishikura class, , is defined as the class of − IS 1 maps P ϕ− : U C, where ϕ : U U is a conformal map with ϕ(0) = 0, ◦ ϕ → → ϕ 2παi ϕ′(0) = e− . They show that there exists a constant N > 0 such that every map in this class with α Irr is infinitely many times renormalizable, and all ∈ N the renormalizations belong to this class. This result has proved to be useful in studying local dynamics of maps with a neutral fixed point. For example, it was used in Buff and Ch´eritat’s proof of existence of positive area Julia sets to control post-critical set of such maps. The idea of the proof of above theorem is to construct infinitely many “gates” with the 0 fixed point on their boundary such that almost every point in the Julia set has to go through them. We can also control diameter of gates in terms of the Brjuno function and conclude that these diameters shrink to zero once α is a non-Brjuno number. The orbit of the critical point goes through all these gates (without any assumption on the area of the Julia set). Therefore, a corollary of the above theorem is the following:
Corollary 1.4. If α is a non-Brjuno number in IrrN (same N as in the above theorem) then every map in the Inou-Shishikura class with the fixed point of multiplier e2παi at 0 is non-linearizable. In particular, if f is a rational map of the form e2παi P h, with α Irr , where h is an arbitrary rational map ◦ ∈ N of the Riemann sphere satisfying
h(0) = 0, h′(0) = 1, • no critical value of h belongs to U (domain of P ), • 9 Then f is not linearizable at 0.
For a rational map of the Riemann sphere f, the post-critical set (f) PC is defined as closure of orbits of all critical points of f. It is proved by Lyu- bich [Ly83b] that the post-critical set of a rational map is the measure theoretic attractor of points in the Julia set of that map. That is, for every neighborhood of the post-critical set, orbit of almost every point in the Julia set eventually stays in that neighborhood. A central result in our work on this class of quadratic maps is the following:
Theorem 1.5. There exists a constant N such that for every non-Bruno α ∈ IrrN , the post-critical set of Pα is connected and has zero area.
A straight corollary of the above theorem is the following:
Corollary 1.6. Almost every point in the Julia set of Pα with a non-Brjuno α Irr is non-recurrent. In particular, there is no finite absolutely continu- ∈ N ous invariant measure on the Julia set.
10 Chapter 2
Preliminaries
2.1 Julia and Fatou sets
Let U be an open subset of the Riemann sphere Cˆ and be a family of holo- F morphic maps defined on U. The family is called normal in U, if every F infinite sequence f ∞ in , contains a subsequence which converges uni- { n}n=1 F formly on all compact subsets of U to a holomorphic map defined on U. Given a rational map f : Cˆ Cˆ, we can consider the family of n-fold compositions → n of f by itself, f ∞ , defined on Cˆ. The Fatou set of f, denoted by F (f), { }n=0 is defined as the largest open subset of Cˆ on which the sequence of iterates
n f ∞ is normal. The Julia set J(f) is defined as the complement of the { }n=0 Fatou set. By definition, these two sets are invariant. The post-critical set of f, which plays an important role in understanding the dynamics of f, is defined as
∞ (f) := f n(c), PC ′ n=1 c:f (c)=0 that is, topological closure of forward images f k(c) with k > 0, of critical points of f. 11 A point z Cˆ is called periodic of period p if f p(z)= z and f j(z) = z for ∈ p 1
2πi p A neutral periodic point is called parabolic if its multiplier is e q for a rational number p/q. Parabolic periodic points belong to the Julia set. A fixed point z with multiplier e2παi is called linearizabale, if there exists a change of coordinate φ defined on some neighborhood of z with φ(z) =0, and
1 2παi φ− f φ = z e z ◦ ◦ → on that neighborhood. If there is no such a coordinate, z is called a Cremer fixed point. In the linearizable case, the maximal domain of linearization is called a Siegel disk. Linearizable periodic points and Cremer periodic points are defined similarly. Every linearizable periodic point belongs to Fatou set and every Cremer one belongs to Julia set. It turns out that the linearizability of a neutral fixed point with an irrational rotation depends on the arithmetic nature of the rotation. Every irrational number can be written as a continued fraction of the form 1 1 , a1 + 1 a2+ . .. which produces best rational approximants
pn 1 = 1 , qn a + 1 1 a2+ ... 1 an 12 for every n 1. The following theorem gives a combinatorial condition on the ≥ rotation which implies linearizability at a neutral fixed point.
Theorem 2.1 (Siegel, Brjuno). If ∞ log q n+1 < , q ∞ n=1 n then every holomorphic germ with a fixed point of multiplier e2παi is locally linearizable.
In the other direction we have,
Theorem 2.2 (Yoccoz). If the above sum is divergent, then the quadratic polynomial z e2παiz + z2 is not linearizable at 0. → By a theorem of Sullivan [Su85] every component of Fatou set is eventually periodic. Let U be a periodic component of Fatou set of a rational map f of period p. There are only five possibilities for f p on U listed below.
1. Supper attracting case: there exists a super attracting periodic point z U, attracting orbit of every point in U under f p. ∈
p 2. Attracting case: there exists a periodic point z U, 0 < (f )′(z) < ∈ | | 1, attracting orbit of every point in U under f p.
p 3. Parabolic case: there exits a periodic point z ∂U with (f )′(z) = 1, ∈ attracting orbit of every point in U under f p.
4. Siegel disk: U is conformally isomorphic to a disk, and f p acts as an irrational rotation on U.
5. Herman Ring: U is conformally isomorphic to an annulus, and f p acts as an irrational rotation on U.
Finally, M. Shishikura [Sh87] proves a conjeture on an upper bound for number of priodic Fatou components. 13 2.2 Dynamics of polynomials
d d 1 Let f : C C be a monic polynomial of degree d, f(z)= z + a z − + + → 1 a , is a super attracting fixed point of f and its basin of attraction is defined d ∞ as D ( )= z C : f n(z) . f ∞ { ∈ → ∞} Its complement is called the filled Julia set; K(f) = C D ( ). The Julia \ f ∞ set J(f) is the common boundary of K(f) and D ( ). It is known that the f ∞ Julia set and the filled Julia set of a polynomial are connected if and only if all critical points stay bounded under iteration. With f as above, there exists a conformal change of coordinate, B¨ottcher coordinate, B which conjugates f to the dth power map z zd throughout f → some neighborhood of infinity Uf . It is defined as,
Bf : Uf z C : z >rf 1 , →{ ∈ | | ≥ } (2.1) dn n Bf (z) := lim f (z) n →∞ n where d th root is chosen such that Bf is tangent to the identity map at infinity.
d By definition, it satisfies the equivariance relation Bf (f(z)) = (Bf (z)) , and B (z) z as z . f ∼ →∞ In particular, if the filled Julia set is connected, Bf coincides with the Riemann mapping of D ( ) onto the complement of the closed unit disk, f ∞ normalized to be tangent to the identity map at infinity.
θ θ The external ray R = Rf of angle θ is defined as
1 iθ B− re : r r r The equipotential E = Ef of level r>rf is defined as 1 iθ B− re :0 θ 2π . f { ≤ ≤ } 14 d It follows from equivariance relation that f(Rθ)= Rdθ, and f(Er)= Er . A ray Rθ is called periodic ray of period p if f p(Rθ) = Rθ. A ray is fixed (p = 1) if and only if θ is a rational number of the form 2πj/(d 1). By − definition, a ray Rθ lands at a well defined point z in J(f) if the limiting value of the ray Rθ (as r r ) exists and equals to z. Such a point z J(f) is → f ∈ called the landing point of Rθ. The following theorem characterizes landing points of periodic rays. See [DH82] for further discussions. Theorem 2.3. Let f be a polynomial of degree d 2 with connected Julia ≥ set. Every periodic ray lands at a well defined periodic point which is either repelling or parabolic. Vice versa, every repelling or parabolic periodic point is the landing point of at least one, and at most finitely many periodic rays with the same ray period. In particular, this theorem implies that the external rays landing at a pe- p 1 riodic point are organized in several cycles. Let a = a − be a repelling or { k}k=0 parabolic cycle of f and let R(ak) be union of all external rays landing at ak. The configuration p 1 − R(a)= (R(ak)) k =0 with the rays labeled by their external angles, is called the periodic point por- trait of f associated to the cycle a. 2.3 Unicritical family and the connectedness locus Any degree d polynomial with only one critical point is affinely conjugate to d Pc(z) = z + c for some complex number c. A case of especial interest is the 15 2πj/(d 1) following fixed point portrait. The d 1 fixed rays R − land at d 1 fixed − − points called βj, and moreover, these are the only rays that land at βj’s. These fixed points are non-dividing, that is, K(F ) β is connected for every j. If the \ j other fixed point, called α, is also repelling, there are at least 2 rays landing at it. Thus, α-fixed point is dividing and by Theorem 2.3, the rays landing at α-fixed point are permuted under the dynamics. The following statement has been shown in [M95] for quadratic polynomials. The same ideas apply to prove it for degree d unicritical polynomials. Proposition 2.4. If at least 2 rays land at the α fixed point of Pc, we have: 1 The component of C P − (R(α)) containing the critical value is a sector • \ c bounded by two external rays. 1 The component of C P − (R(α)) containing the critical point is a region • \ c bounded by 2d external rays landing in pairs at the points e2πj/dα, for j =0, 1,...,d 1. − The Connectedness locus of degree d is defined as the set of parameters Md c in C for which J(Pc) is connected, or equivalently, the critical point of Pc does not escape to infinity under iteration of P . In particular, is the well- c M2 known Mandelbrot set. See Figures 2.1 and 2.2. A well-known result due to Douady and Hubbard [DH82] shows that the connectedness loci are connected. Their argument is based on constructing an explicit conformal isomorphism B : C d z C : z > 1 Md \M →{ ∈ | | } given by B (c)= Bc(c), where Bc is the B¨ottcher coordinate for Pc. Md By means of conformal isomorphism B , the parameter external rays Md Rθ and equipotentials Er are defined, similarly, as the B -preimages of the Md straight rays going to infinity and round circles around 0. 16 Figure 2.1: The Mandelbrot set A polynomial Pc (and the corresponding parameter c) is called hyperbolic if P has an attracting periodic point. The set of hyperbolic parameters in , c Md topologically open by definition, is the union of some components of int . Md These components are called hyperbolic components. The main hyperbolic component is defined as the set of parameter values c for which Pc has an attracting fixed point. Outside of the closure of this set all fixed points become repelling. Now, consider a hyperbolic component H ⊂ int , and suppose b is the corresponding attracting cycle with period k. On Md c the boundary of H this cycle becomes neutral, and there are d 1 parameters − c ∂H where P has a parabolic cycle with multiplier equal to one. One of i ∈ ci these parameters ci, denoted by croot, which divides the connectedness locus into two pieces, is called root of H (See [DH82] for quadratic polynomials and [Sch04] for arbitrary degree unicritical polynomials). Indeed, any hyperbolic component has one root and d 2 co-roots. The root is the landing point of two − parameter rays, while every co-root is the landing point of a single parameter ray, See Figure 2.3. 17 Figure 2.2: Figure on the left shows the connectedness locus . The figure M3 on the right is an enlargement of a primary limb in . The dark regions M3 show some of the secondary limbs Let c belong to a hyperbolic component H , different from main component ¯ of the connectedness locus, with attracting cycle bc. The basin of attraction of ¯b , denoted by A , is defined as the set of points z C with P n(z) converges c c ∈ c to the cycle bc. The boundary of the component of Ac containing c is a jordan k curve which we denote it by Dc. The map Pc on Dc is topologically conjugate to θ dθ on the unit circle. Therefore, there are d 1 fixed points of P k on → − c this jordan curve which are repelling periodic points (of Pc) of period dividing period of ¯bc (its period can be strictly less than period of bc). Among all rays landing at these repelling periodic points, let θ1 and θ2 be the angles of the external rays bounding the sector containing the critical value of Pc (See Figure 2.3). The following theorem makes a connection between external rays Rθ1 , θ2 R and the parameter external rays Rθ1 , Rθ2 on the parameter plane. See [DH82] or [Sch04] for proofs. 18 Rθ3 -ray landing at a co-root Rθ1 Rθ2 Rθ1 -ray landing at a root Rθ2 -ray landing at a root Figure 2.3: Figure on the left shows a primitive renormalizable Julia set and the external rays Rθ1 and Rθ2 landing at the corresponding repelling periodic point. The figure on the right is the corresponding primitive little multibrot copy. It also shows the parameter external rays Rθ1 and Rθ2 landing at the root point. Theorem 2.5. The parameter external rays Rθ1 and Rθ2 land at the root point of H , and these are the only rays that land at this point. Closure of the two parameter external rays Rθ1 and Rθ2 cut the plane into two components. The one containing the component H , with the root point attached to it, is called the wake WH . So, a wake is an open set with a root point attached to its boundary. For a wake WH and an equipotential of level η, E , the truncated wake WH (η) is the bounded component of WH E . η \ η Part of the connectedness locus contained in the wake WH with the root point attached to it is called the limb LH of the connectedness locus originated at H . In other words, the limb LH is part of contained in WH . By Md 19 definition, every limb is a closed set. The wakes attached to the main hyperbolic component of are called Md primary wakes and a limb associated to such a primary wake will be called primary limb. If H is a hyperbolic component attached to the main hyperbolic component, all the wakes attached to such a component H (except WH itself) are called secondary wakes. Similarly, a limb associated to a secondary wake will be called secondary limb. A truncated limb is obtained from a limb by removing a neighborhood of its root. Some secondary limbs are shown in Figure 2.2. Given a parameter c in H , we have the attracting cycle bc as above, and the θ1 associated repelling cycle ac which is the landing point of the external rays R and Rθ2 . The following result gives the dynamical meaning of the parameter values in the wake WH which is bounded by parameter external rays Rθ1 and Rθ2 (See [Sch04] for the proof). Theorem 2.6. For parameters c in WH root point , the repelling cycle a \{ } c stays repelling and moreover, the isotopic type of the ray portrait R(ac) is fixed throughout WH . 2.4 Quasi-conformal mappings There are several equivalent definitions of quasi-conformal mappings conve- nient in different situations. A good source on the subject is Ahlfor’s book [Ah66]. An analytic definition is more appropriate for us. Let Ω be an open subset of the complex plane. A homeomorphism φ : Ω C is called absolutely → continuous on lines if for every rectangle R contained in Ω with sides parallel to x and y axis, φ is absolutely continuous on almost every horizontal and almost every vertical line in R. Such function has partial derivatives almost 20 every where in Ω. Let φ denote 1 (φ iφ ), and φ denote 1 (φ + iφ ), with z 2 x − y z¯ 2 x y z = x + iy. An orientation preserving homeomorphism φ : Ω C is called K-quasi- → conformal, K-q.c. for short, if φ is absolutely continuous on lines, • K 1 φ k φ almost every where in Ω, where k = − . • | z¯| ≤ | z| K+1 The measurable function φ = φz¯/φz is called complex dilatation of φ and the smallest value of K satisfying above definition is called dilatation of φ. Some properties of q.c. mappings which will be used in our work is listed in the following theorem. One may refer to [Ah66] for their proofs. Theorem 2.7. Quasi-conformal mappings satisfy the following properties: Composition of a K -q.c. and a K -q.c. map is a K K -q.c. map. • 1 2 1 2 The space of K-q.c. mappings from Cˆ to Cˆ fixing three points 0, 1, and • is a compact class (under uniform convergence on compact sets). ∞ A 1-q.c. map is conformal. • We will be interested in solving Beltrami equation φ = for a given measurable function : Cˆ Cˆ. Solution to this problem is referred to as → measurable Riemann mapping theorem. Theorem 2.8. For any measurable : Cˆ Cˆ with < 1, there exists a → ∞ unique normalized q.c. mapping φ with complex dilatation that leaves 0, 1, fixed. ∞ Any topological annulus is conformally isomorphic to one of C 0 , B(0, 1) \{ } \ 0 , or B(0,r) B(0, 1) for some r > 1. It’s modulus, denoted as mod A, is { } \ 21 defined as in the first two cases and 1 log r in the last case. It turns out ∞ 2π that this quantity is quasi-invariant under q.c. mappings. That is, given a K-q.c. mapping φ : A C, we have 1 mod A mod φ(A) K mod A. → K ≤ ≤ The following theorem makes the connection between q.c. mappings on Ω and maps on boundary of Ω. For convenience we will assume that Ω is upper half plane. Theorem 2.9 (boundary correspondence). Let φ be a K-q.c. mapping of upper half plane onto itself which maps to itself. Then φ induces a homeomorphism ∞ h : R R which satisfies → 1 h(x + t) h(x) M(K)− − M(K), ≤ h(x) h(x t) ≤ − − for some constant M(K) depending only on K. Vice versa, every homeomor- phism h : R R with h( ) = that satisfies above inequality for some M, → ∞ ∞ extends to a q.c. mapping of the upper half plane with dilatation depending only on M. 2.5 Polynomial-like maps A holomorphic proper branched covering of degree d, f : U ′ U, where U and → U ′ are simply connected domains with U ′ compactly contained in U, is called a polynomial-like map. Reader can consult [DH85] for the following material on polynomial-like maps. Every polynomial can be viewed as a polynomial-like map once restricted to an appropriate neighborhood of its filled Julia set. In what follows we will only consider polynomial-like maps with one branched point of degree d (which is assumed to be at zero after normalization) and refer to them as unicritical polynomial-like maps. 22 The filled Julia set K(f) of a polynomial-like map f : U ′ U is naturally → defined as n K(f)= z C : f (z) U ′, n =0, 1, 2,... . { ∈ ∈ } The Julia set J(f) is defined as the boundary of K(f). They are connected if and only if K(f) contains critical point of f. Two polynomial-like maps f and g are said topologically (quasi-conformally, conformally, affinely) conjugate if there is a choice of domains f : U ′ U and → g : V ′ V and a homeomorphism h : U V (quasi-conformal, conformal, or → → affine isomorphism, respectively) such that h f U ′ = g h U ′. ◦ | ◦ | Two polynomial like maps f and g are hybrid or internally equivalent if there is a q.c. conjugacy h between f and g with ∂h = 0 on K(f). The following theorem due to Douady and Hubbard [DH85] makes the connection between polynomial-like maps and polynomials. Theorem 2.10 (Straightening). Every polynomial-like map f is hybrid equiv- alent to (suitable restriction of) a polynomial P of the same degree. Moreover, P is unique up to affine conjugacy when K(f) is connected. In particular, any unicritical polynomial-like map with connected Julia set corresponds to a unique (up to affine conjugacy) unicritical polynomial z → zd + c with c in the connectedness locus . Note that zd + c and zd + c/λ Md are conjugate via z λz for every d 1th root of unity λ. → − Given a polynomial-like map f : U ′ U, we can consider the fundamental → annulus A = U U ′. It is not canonic because any choice of V ′ ⋐ V such \ that f : V ′ V is a polynomial-like map with the same Julia set will give a → different annulus. However we can associate a real number, modulus of f, to any polynomial-like map f as follows: mod(f) = sup mod(A) 23 where the sup is taken over all possible fundamental annuli A of f. It can be seen from the proof of the straightening theorem that dilatation of the q.c. conjugacy between f and the corresponding polynomial P : z c(f) → zd + c(f) can be controlled by modulus of f. Proposition 2.11. If mod(f) > 0 then dilatation of the q.c. conjugacy ≥ obtained in the straightening theorem depends only on . 2.6 Holomorphic motions In this section we consider analytic deformations of sets in Cˆ. Let A be a subset of Cˆ. A holomorphic motion of A parametrized on a complex manifold M is a map Φ : M A Cˆ such that × → For any fixed z A, the map λ Φ(λ, z) is holomorphic in M. • ∈ → For any fixed λ M, the map z Φ(λ, z) is an injection. • ∈ → There exists a point λ M with Φ(λ , z)= z, for every z A. • 0 ∈ 0 ∈ The following remarkable result due to Man´e-Sad-Sullivan [MSS83] relates holomorphic dynamics with q.c. mappings. Theorem 2.12 (λ-Lemma). If Φ: M A Cˆ is a holomorphic motion, then × → Φ has an extension to Φ:˜ M A¯ Cˆ, with × → Φ˜ is a holomorphic motion of A¯. • Each Φ(˜ λ, ): A¯ Cˆ is quasi-conformal. • → An important result about holomorphic motions is Slodkowski’s generalized λ-lemma [Sl91] stating that a holomorphic motion of a set A extends to a holomorphic motion of the whole Riemann sphere. 24 Chapter 3 Combinatorial Rigidity in the Unicritical Family 3.1 Modified principal nest 3.1.1 Yoccoz puzzle pieces Recall that for a parameter c outside of the main component of the ∈ Md connectedness locus, P has a unique dividing fixed point α . The q 2 c c ≥ external rays R(αc) landing at this fixed point together with an arbitrary equipotential Er, cut the domain inside Er into q closed topological disks Y 0, j = 0, 1,...,q 1, called puzzle pieces of level zero. That is, Y 0’s are the j − j closures of the bounded components of C Er R(¯α ) α . The main \{ ∪ c ∪{ c}} 0 property of this partition is that Pc(∂Yj ) does not intersect interior of any 0 piece Yi . n Now the puzzle pieces Yi of level or depth n are defined as the closures of n 0 the connected components of f − (int(Yj )). They partition the neighborhood n r of the filled Julia set bounded by equipotential f − (E ) into finite number of 25 closed disks. By definition all puzzle pieces are bounded by piecewise analytic curves. The label of each puzzle piece is the set of the angles of external rays bounding that puzzle piece. If the critical point does not land on the αc-fixed n point, there is a unique puzzle piece Y0 of level n containing the critical point. The family of all puzzle pieces of Pc of all levels has the following Markov property: Puzzle pieces are disjoint or nested. In the latter case, the puzzle piece • of higher level is contain in the puzzle piece of lower level. Image of any puzzle piece of level n 1 is a puzzle piece of level n 1. • − n n 1 Moreover, P : Y Y − is d-to-1 branched covering or univalent, c j → k n depending on whether Yj contains the critical point or not. On the first level, there are d(q 1)+1 puzzle pieces. One critical piece Y 1, − 0 q 1 ones, denoted by Y 1, attached to the fixed point α , and the (d 1)(q 1) − i c − − 1 1 1 symmetric ones, denoted by Z , attached to P − (α ) α . Moreover f Y i c c \{ c} | 0 1 1 1 1 d-to-1 covers Y1 , f Yi univalently covers Yi+1, for i =1,...,q 2, and f Yq 1 | − | − 1 (d 1)(q 1) 1 q 1 1 r univalently covers, Y − − Z . Thus f (Y ) truncated by f − (E ) is 0 ∪ i=1 i 0 1 1 the union of Y0 and Zi ’s. We will assume after this that P n(0) = α for all n, so that the critical c c puzzle pieces of all levels are well defined. As it will be apparent in a moment, this condition is always the case for renormalizable polynomials. 3.1.2 The complex bounds in the favorite nest and renor- malization For a puzzle piece V 0, let R : Dom R V V denote the first return ∋ V V ⊆ → map to V . It is defined at every points z in V for which there exists a positive 26 integer t with P t(z) int V . Then R (z) is defined as P t(z) where t is the c ∈ V c first positive moment when P t(z) int V . Markov property of puzzle pieces c ∈ implies that any component of Dom RV is contained in V and the restriction of t this return map (Pc , for some t) to such a component is d-to-1 or 1-to-1 proper map onto V . The component of the Dom RV which contains the critical point is called the central component of RV . If the image of critical point under the first return map belongs to the central component, the return is called central return. The first landing map L to a puzzle piece V 0 is defined at all points V ∋ z C for which there exists an integer t 0 with P t(z) int V . It is the ∈ ≥ c ∈ identity map on the component V and univalently maps each component of Dom LV onto V . Consider a puzzle piece Q 0. If the critical point returns back to Q under ∋ iteration of P , the central component P Q of R is the pullback of Q by P p c ⊂ Q c along the orbit of the critical point, where p is the first moment when critical orbit enters int Q. This puzzle piece P is called the first child of Q. Recall that P p : P Q is a proper map of degree d. c → The favorite child Q′ of Q is constructed as follows; Let p> 0 be the first moment when Rp (0) int(Q P ) (if it exists) and q > 0 be the first moment Q ∈ \ (if it exists) when Rp+q(0) int P (p + q is the moment of the first return ∈ back to P after the first escape of the critical point from P under iterates of p+q RQ). Now Q′ is defined as the pullback of Q under RQ containing the critical k p+q point. Markov property of puzzle pieces implies that the map Pc = RQ (for an appropriate k > 0) from Q′ to Q is proper of degree d. The main property of the favorite child is that the image of the critical point under the k map P : Q′ Q belongs to the first child P . c → 27 Consider a unicritical polynomial Pc with q rays landing at its α fixed point and form the corresponding Yoccoz puzzle pieces. The map Pc is called satellite renormalizable, or immediately renormalizable if P lq(0) Y 1, for l =0, 1, 2,.... c ∈ 0 The map P q : Y 1 P q(Y 1) is proper of degree d but its domain is not c 0 → c 0 1 1 compactly contained in its range. By slight “thickening” of Y0 so that Y0 is q 1 q compactly contained in Pc (Y0 ) (see [M00]), Pc can be turned into a unicritical polynomial-like map. Note that the above condition implies that the critical point does not escape its domain of definition and therefore the corresponding little Julia set is connected. If Pc is not satellite renormalizable, then there is a first moment k such kq 1 1 1 that Pc (0) belongs to some Zi . Define Q as the pullback of this Zi under kq 1 Pc . By the above construction we form the first child, P , and the favorite child ,Q2, of Q1. Repeating the above process we get a nest of puzzle pieces Q1 P 1 Q2 P 2 Qn P n (3.1) ⊃ ⊃ ⊃ ⊃ ⊃ ⊃ ⊃ where P i is the first child of Qi, and Qi+1 is the favorite child of Qi. The above process stops if and only if one of the following happens: The map P is combinatorially non-recurrent, that is, the critical point • c does not return to some critical puzzle piece. n The critical point does not escape the first child P under iterates of R n • Q for some n, or equivalently, returns to all critical puzzle pieces of level bigger than n are central. In the former case, combinatorial rigidity in the critically non-recurrent case has been taken care of in [M00]. 28 k n n In the latter case, R n = P : P Q (for an appropriate k) is a Q c → unicritical polynomial-like map of degree d with P n compactly contained in Qn. The map P is called primitively renormalizable in this case. Note that the corresponding little Julia set is connected because all the returns of critical point to Qn are central by definition. A map P : C C is called renormalizable if it is satellite or primitively c → renormalizable. To deal with non-renormalizable and recurrent polynomials, the following a priori bounds has been proved in [AKLS05]. Theorem 3.1. There exists δ > 0 such that for every ε > 0 there exists n0 = n0(ε) > 0 with the following property. For the nest of puzzle pieces Q1 P 1 Q2 P 2 Qm P m ⊃ ⊃ ⊃ ⊃ ⊃ ⊃ ⊃ as above, if mod (Q1 P 1) >ε and n If a map Pc is combinatorially recurrent, the critical point does not land at α-fixed point, therefore puzzle pieces of all levels are well defined. The combinatorics of Pc up to level n is an equivalence relation on the set of angles of puzzle pieces up to level n. Two angles θ1 and θ2 are equivalent if the corresponding rays Rθ1 and Rθ2 land at the same point. One can see that the n combinatorics of a map up to level n+t determines the puzzle piece Yj of level n containing the critical value f t(0). Two non-renormalizable maps are called combinatorially equivalent if they have the same combinatorics up to arbitrary level n,that is, they have the same set of labels of puzzle pieces and the same equivalence relation on them. Combinatorics of a renormalizable map will be defined in section 3.1.3. Two unicritical polynomials Pc and Pc˜ with the same combinatorics up to level n are called pseudo-conjugate (up to level n) if there is an orientation 29 preserving homeomorphism H : (C, 0) (C, 0), such that H(Y 0) = Y 0 for → j j j =0, 1, 2,... and H P = P H outside of the critical puzzle piece Y n. A ◦ c c˜ ◦ 0 pseudo-conjugacy H is said to match the B¨ottcher marking if near infinity it becomes identity in the B¨ottcher coordinates for Pc and Pc˜. Thus a pseudo- conjugacy is the identity map in the B¨ottcher coordinate outside of Y n by ∪j j its equivariance property. m m Let qm and pm be the levels of the puzzle pieces Q and P , that is, m qm m pm Q = Y0 and P = Y0 . The following statement is the main technical result of [AKLS05] which will be used frequently in our construction. Theorem 3.2. Assume that a (finite) nest of puzzle pieces Q1 P 1 Q2 P 2 Qm P m (3.2) ⊃ ⊃ ⊃ ⊃ ⊃ ⊃ is obtained for Pc. If Pc˜ is combinatorially equivalent to Pc, then there exists a m qm K-q.c. pseudo-conjugacy H (up to level qm where Q = Y0 ) between Pc and Pc˜ which matches the B¨ottcher marking. Proposition 3.3. Assume that the nest of puzzle pieces in the above theorem is defined using equipotential of level η, then dilatation of the q.c. pseudo- conjugacy, K = K(c, c˜), obtained in that theorem depends only on the distance between c and c˜ in the primary wake truncated by equipotential of level η and equipped with the Poincar´emetric. Proof. To prove the proposition we need a brief sketch of the proof of the above theorem. For more details refer to [AKLS05]. Combinatorial equivalence of Pc and Pc˜ up to level zero implies that the parameters c andc ˜ belong to the same truncated wake W (η) attached to the main component of the Connectedness locus. Inside W (η), the q external rays R(α) and the equipotential E(h) (for 30 every h > η) move holomorphically in C 0. That is, there exists a holomor- \ 1 phic motion Φ of R(α) E(h), given by B− B in the second coordinate, ∪ c˜ ◦ c parametrized on W (η) such that φ(˜c, R(α) E(h)) = (˜c, R(α) E(h)). ∪ ∪ Outside of equipotential E(h), this holomorphic motion extends to a mo- tion holomorphic in both variables (c, z) which is coming from the B¨ottcher c˜ c 1 coordinate near . By [Sl91] the map φ (φ )− extends to a K q.c. ∞ ◦ 0 map G : (C, 0) (C, 0), where K only depends on the hyperbolic dis- 0 → 0 tance between c andc ˜ in the truncated wake W (η). This gives a q.c. map G : (C, 0) (C, 0) which conjugates P and P outside of puzzle pieces of 0 → c c˜ level zero. By adjusting the q.c. map G0 inside equipotential E(h) such that it sends c toc ˜, we get a q.c. map (not necessarily with the same dilatation) G0′ . By ˜ lifting G0′ via Pc and f we get a new q.c. map G1. Repeating this process, for i = 1, 2,...,n = qm, which is adjusting the q.c. map Gi inside union of puzzle pieces of level i + 1 so that it sends c toc ˜ and lifting it, we obtain a q.c. map Gi+1 (not with the same dilatation) conjugating Pc and Pc˜ outside union of puzzle pieces of level i + 1. At the end we will have a q.c. map Gn which n conjugates Pc and Pc˜ outside of equipotential E(h/d ). The nest of puzzle pieces Q1 P 1 Q2 P 2 Qm P m ⊃ ⊃ ⊃ ⊃ ⊃ ⊇ for Pc˜ is defined as the image of the nest of puzzle pieces in (3.2) under the map Gn. Combinatorial equivalence of Pc and Pc˜ implies that this new nest i+1 has the same properties as the one for Pc. In other words, Q is the favorite child of Qi and P i is the first child of Qi. Hence Theorem 3.1, applies to this 31 nest too. By properties of these nests, one constructs a K-q.c. map Hn from the critical puzzle piece Qn to the corresponding one Qn where K only depends on the a priori bounds δ and the hyperbolic distance between c andc ˜ in the η truncated wake W . The pseudo-conjugacy Hn is obtained from univalent lifts of Hn onto other puzzle pieces. If Pc is renormalizable, the process of constructing modified principal nest stops at some level and returns to the critical puzzle pieces after this level are central. One can see that critical puzzle pieces do not shrink to 0. 3.1.3 Combinatorics of a map If P : z zd +c is renormalizable, there is a homeomorphic copy 1 c of c0 → 0 Md ∋ 0 the connectedness locus within the connectedness locus satisfying the following properties (see[DH85]): For c 1 the root point , P : z zd + c is ∈ Md \{ } c → renormalizable, and there is a holomorphic motion of the dividing fixed point α and the rays landing at it on a neighborhood of 1 the root point , such c Md \{ } that the renormalization of Pc is associated to this fixed point and external rays. In other words, all parameters in this copy have Yoccoz puzzle pieces of all levels with the same labels. This homeomorphism is not unique because of the symmetry in the connectedness locus. However, we define it uniquely by sending the unique root point of the copy to the landing point of the parameter external ray of angle 0. We show below that among all renormalizations of Pc, there is a unique one denoted by P which corresponds to a maximal copy (not included in any R c other copy except itself) of the connectedness locus inside the connected- Md ness locus. j Assume that P is equal to P : U U ′, for some positive integer j and R c c → 32 topological disk U compactly contained in U ′. By straightening theorem, P R c is conjugate to a unicritical polynomial Pc′ . That is, there exists a q.c. mapping j χ : U ′ C, with χ P (z) = P ′ χ(z), for every z U. This polynomial → ◦ c c ◦ ∈ Pc′ is determined up to conformal equivalence in the theorem. However, there are only d 1 conformally equivalent polynomials in each class. We make − this parameter unique by choosing the image of c under the homeomorphism uniquely determined above. If Pc′ is also renormalizable, Pc is called twice renormalizable. Let positive k integer k, and topological disks V and V ′ be such that P ′ : V V ′ is a renor- c → malization of Pc′ . Define V and V ′ as χ-preimage of V and V ′, respectively. jk k One can see that χ conjugates P : V V ′ with P ′ : V V ′. Therefore, c → c → jk 2 P : V V ′ is also a polynomial-like map. We denote this map by P . c → R c Above process may be continued to associate a canonical finite or infinite sequence P , P , 2P ,..., of polynomial-like maps to P , and accordingly, c R c R c c call Pc at most finitely, or infinitely renormalizable. Equivalently, there is a finite or infinite sequence τ(P ) := 1, 2,... , of maximal connectedness c Md Md locus copies associated to P , where n corresponds to the renormalization c Md n n 1 P of − P . In the infinitely renormalizable situation, τ(P ) is called the R c R c c combinatorics of Pc. Two infinitely renormalizable maps are called combinatorially equivalent if they have the same combinatorics, i.e. correspond to the same sequence of maximal connectedness locus copies. We say an infinitely renormalizable map Pc satisfies the secondary limbs condition, if all of the connectedness copies in the combinatorics τ(Pc) belong to a finite number of truncated secondary limbs. Let stand for the class SL of infinitely unicritical polynomial-like maps satisfying the secondary limbs 33 condition. An infinitely renormalizable map Pc has a priori bounds, if there is an ε> 0 with mod ( mP ) ε, for all m 0. R c ≥ ≥ 34 3.2 Proof of the rigidity theorem In this section we begin to prove the rigidity theorem in a slightly more general form as follows. 3.2.1 Reductions Theorem 3.4 (Rigidity theorem). Let f and f˜ be two infinitely renormalizable unicritical polynomial-like maps satisfying condition with a priori bounds. SL If f and f˜ are combinatorially equivalent, then they are hybrid equivalent. Remark. If the two maps f and f˜ in the above theorem are polynomials, then hybrid equivalence becomes conformal equivalence. That is because the B¨ottcher coordinate, which conformally conjugates the two maps on the com- plement of the Julia sets, can be glued to the hybrid conjugacy on the Julia set. See Proposition 6 in [DH85] for a precise proof of this. The proof breaks into following steps: combinatorial equivalence ⇓ topological equivalence ⇓ q.c. equivalence ⇓ hybrid equivalence It has been shown in [J00] that any unbranched infinitely renormalizable map with a priori bounds has locally connected Julia set. A renormalization 35 f n : U V is called unbranched if (f) U = (f n : U V ). Here, → PC ∩ PC → unbranched condition follows from our combinatorial condition and a priori bounds (see [Ly97] Lemma 9.3). Then the first step, topological equivalence of combinatorially equivalent maps, follows from the local connectivity of the Julia sets by the Carath´eodory theorem. Indeed by [Do93] there is a topological model for the Julia set of these maps based on their combinatorics. The last step in general follows from McMullen’s Rigidity Theorem [Mc94] (Theorem 10.2). He has shown that an infinitely renormalizable quadratic polynomial-like map with a priori bounds does not have any nontrivial invari- ant line field on its Julia set. The same proof works for degree d unicritical polynomial-like maps. It follows that any q.c. conjugacy h between f and f˜ satisfies ∂h = 0 almost everywhere on the Julia set. Therefore, h is a hybrid conjugacy between f and f˜. However, if all infinitely renormalizable unicrit- ical maps in a given combinatorial class satisfy a priori bounds condition, it is easier to show that q.c. conjugacy implies hybrid conjugacy for that class rather than showing that there is no nontrivial invariant line field on the Julia set. Since we are finally going to apply our theorem to combinatorial classes for which a priori bounds have been established, we will prove it in Proposi- tion 3.20. So assume that f and f˜ are topologically conjugate. We want to show the following: Theorem 3.5. Let f and f˜ be infinitely renormalizable unicritical polynomial- like maps satisfying a priori bounds and conditions. If f and f˜ are topo- SL logically conjugate then they are q.c. conjugate. 36 3.2.2 Thurston equivalence Suppose two unicritical polynomial-like maps f : U U and f˜ : U U 2 → 1 2 → 1 are topologically conjugate. A q.c. map h :(U , U , (f)) (U , U , (f˜)) 1 2 PC → 1 2 PC is a Thurston conjugacy if it is homotopic to a topological conjugacy ψ :(U , U , (f)) (U , U , (f˜)) 1 2 PC → 1 2 PC between f and f˜ relative ∂U ∂U (f). Note that a Thurston conjugacy 1 ∪ 2 ∪ PC is not a conjugacy between two maps. It is a conjugacy on the postcritical set and homotopic to a conjugacy on the complement of the postcritical set. We will see in the next lemma that it is in the “right” homotopy class. The following result is due to Thurston and Sullivan [S92] which originates the “Pull-Back Method” in holomorphic dynamics. Lemma 3.6. Thurston conjugate unicritical polynomial-like maps are q.c. con- jugate. Proof. Assume h :(U , U , (f)) (U , U , (f˜)) is a Thurston conjugacy 1 1 2 PC → 1 2 PC homotopic to a topological conjugacy Ψ : (U , U , (f)) (U , U , (f˜)) 1 2 PC → 1 2 PC relative ∂U ∂U (f). 1 ∪ 2 ∪ PC As f : U 0 U f(0) and f˜ : U 0 U f˜(0) are covering 2 \{ }→ 1 \{ } 2 \{ }→ 1 \{ } maps, h : U f(0) U f˜(0) can be lifted to a homeomorphism 1 1 \ { } → 1 \ { } h : U 0 U 0 . Moreover, since h satisfies the equivariance relation 2 2 \{ }→ 2 \{ } 1 h f = f˜ h on the boundary of U , h can be extended onto U U by h . It 1 ◦ ◦ 1 2 2 1 \ 2 1 also extends to the critical point by sending it to the critical point of f˜. Let us denote this new map by h2. For the same reason, every homotopy ht between 37 ˜ Ψ and h1 can be lifted to a homotopy between Ψ and h2. As f and f are holomorphic maps, h2 has the same dilatation as dilatation of h1. This implies that the new map h2 is also a Thurston conjugacy with the same dilatation as the one of h1. By definition, the new map h2 satisfies the equivariance relation 1 on the annulus U f − (U ). 2 \ 2 Repeating the same process with h2, we obtain a q.c. map h3 and so on. Thus, we have a sequence of K-q.c. maps hn from U1 to U1 which satisfies n equivariance relation on the annulus U f − (U ). All these maps can be 2 \ 2 extended onto complex plane (see Theorem 2.9) with a uniform bound on their dilatation. This family of q.c. maps is normalized at points 0, f(0), f 2(0),... by mapping them to the corresponding points 0, f˜(0), f˜2(0),... . Compactness of this class, Theorem 2.7, implies that there is a subsequence hnj which converges to a K-q.c. map H on U1. For every z outside of the Julia set, the sequence hnj (z) stabilizes and, by definition, eventually h f(z) = f h (z). Taking limit of both sides nj ◦ ◦ nj will imply that H f(z)= f H(z) for every such z. As filled Julia set of an ◦ ◦ infinitely renormalizable unicritical map has empty interior, conjugacy relation for an arbitrary z on the Julia set follows from continuity of H. By the a priori bounds assumption in the theorem, there are topological disks V ⋐ U containing 0 such that nf = f tn : V U is a unicritical n,0 n,0 R n,0 → n,0 degree d polynomial-like map and mod (U V ) ε. By going several levels n,0 \ n,0 ≥ tn ktn ktn down, i.e. considering f : f − (V ) f − (U ) for some positive integer n → n k, we may assume that mod (U V ) and mod (U V ) are proportional. n \ n n \ n Also by slightly shrinking the domains, if necessary, we may assume that these domains have smooth boundaries. Thus we have - For every n 1, we have mod (U V ) ε,and mod (U V ) ε, ≥ n,0\ n,0 ≥ n,0 \ n,0 ≥ 38 - There exits a constant M such that for every n 1, ≥ 1 mod (U V ) n \ n M, M ≤ mod (U V ) ≤ n \ n - For every n 1, U and U have smooth boundaries. ≥ n,0 n,0 We use the following notations throughout the rest of this note. f : V U ,K = K(f), 0 → 0 0,0 f = f t1 : V U ,K = K( f), R 1,0 → 1,0 1,0 R 2f = f t2 : V U ,K = K( 2f), R 2,0 → 2,0 2,0 R . . nf = f tn : V U ,K = K( nf), R n,0 → n,0 n,0 R . . The domain V , for i = 1, 2,...,t 1, is defined as the pullback of n,i n − i tn i Vn,0 under f − containing the little Julia set Jn,i := f − (Jn,0) and Un,i as the i tn component of f − (U ) containing V so that f : V U is a polynomial- n,0 n,i n,i → n,i like map. The domain Wn,i is defined as the preimage of Vn,i under the map f tn : V U . n,i → n,i i Accordingly, Kn,i is defined as the component of f − (Kn,0) inside Vn,i. Note that nf : V U is a polynomial-like map with the filled Julia R n,i → n,i set K which is conjugate to nf : V U by conformal isomorphism n,i R n,0 → n,0 f i : U U . It has been proved in [Ly97] (Lemma 9.2) that there is al- n,i → n,0 ways definite space in between Julia sets in the primitive case for parameters under our assumption. Compare the proof of Lemma 3.13. It has been shown in [Mc96] that definite space between little Julia sets implies that there exist choice of domains Un,i which are disjoint for different i’s and moreover the annuli U V have definite moduli. So we will assume that on the primitive n,i \ n,i levels, the domains Un,i are disjoint for different i’s. 39 In all of the above notation, the first lower subscripts denote the level of renormalization and the second lower subscripts run over little filled Julia sets, Julia sets and their neighborhoods accordingly. In what follows all correspond- ing objects for f˜ will be marked with a tilde and any notation introduced for f will be automatically introduced for f˜ too. To build a Thurston conjugacy, we first introduce multiply connected do- mains Ωn(k),i (and Ωn(k),i) in C for an appropriate subsequence n(k) of the renormalization levels and a sequence of q.c. maps with uniformly bounded distortion h : Ω Ω n(k),i n(k),i → n(k),i for k =0, 1, 2,... and i =0, 1, 2,...,t 1. These domains will satisfy the n(k) − following properties: Each Ω , for n(k) 0, is a topological disk minus n(k+1) topological • n(k),j ≥ disks Dn(k+1),i. Each Ω , for n(k) 1, is well inside D which means that the mod- • n(k),i ≥ n(k),i ulus of the annulus obtained from D Ω is uniformly bounded n(k),i \ n(k),i below for all n(k) and i. Every little postcritical set J (f) is well inside D . • n(k),i ∩ PC n(k),i i Every D is the pullback of D under f − containing J (f) • n(k),i n(k),0 n(k),i∩PC i and every Ωn(k),i is the component of f − (Ωn(k),0) inside Dn(k),i. Finally, we construct a Thurston conjugacy by an appropriate gluing of the maps h : Ω Ω together on the complement of all these multiply n(k),i n(k),i → n(k),i connected domains (which is a union of annuli). See Figure 3.1. 40 Dl,0 Ωl,0 Dl+1,0 Ωl+1,0 D˜ l,0 Ω˜ l,0 D˜ l+1,0 Ω˜ l+1,0 b b b b gl b b hl,0 gl+1 hl+1,0 Figure 3.1: The multiply connected domains and the buffers 3.2.3 The domains Ωn,j and the maps hn,j By applying the straightening theorem to the polynomial-like map n 1 − f : Vn 1,0 Un 1,0 R − → − we get a K1(ε)-q.c. map and a unicritical polynomial fcn−1 , such that 0 1 Sn 1 :(Un 1,0,Vn 1,0, 0) (Υn 1, Υn 1, 0), (3.3) − − − → − − n 1 Sn 1 − f = fcn−1 Sn 1. − ◦ R ◦ − See Figure 3.2. Remark. To make notations easier to follow, we will drop the second subscript whenever it is zero and it does not create any confusion. Also, all objects on the dynamic planes of fcn−1 and fc˜n−1 (the ones after straightening) will be denoted by bold face of notations used for objects on the dynamic planes of f and f˜. To define Ωn 1,j and hn 1,j, because of of the difference in type of renor- − − malizations, we will consider the following three cases: 41 n 1 A . − f is primitively renormalizable. R n 1 n B. − f is satellite renormalizable and f is primitively renormalizable. R R n 1 n C . − f is satellite renormalizable and f is also satellite renormalizable. R R For a given infinitely renormalizable map fc, the renormalization on each level is of primitive or satellite type. Therefore, we can associate a word P...PS...SP... of P and S where a P or a S in the i’s place means that the i’s renormaliza- tion of fc is of primitive or satellite type, respectively. Corresponding to any m1 m2 m3 such word, we define a word of cases A B C ... (for non-negative mj’s) defined as follows. Starting from left, a P is replaced by A , SP by B, and SS by C S. By repeating this process, we obtain a word of cases which is used to decide which case to pick at each step. Case A : We need the following lemma to show that there are equipotentials of suffi- ciently high level η(ε) inside Sn 1(Wn 1,0) and S˜n 1(Wn 1,0) in the dynamic − − − − planes of the maps f − and f − . cn 1 c˜n 1 Lemma 3.7. If P : U ′ U is a unicritical polynomial with connected Julia c → set and mod (U U ′) ε, then U ′ contains equipotentials of level less than \ ≥ η(ε) depending only on ε. Proof. The map Pc on the complement of K(Pc) is conjugate to P0 on the complement of the closed unit disk D1 by B¨ottcher coordinate Bc. Since levels of equipotentials are preserved under this map and modulus is conformal in- variant, it is enough to prove the statement for P : V ′ V for V ′ compactly 0 → 42 Rn−1f n−1 ˜ hn−1,0 R f n−2 R f Rn−2f˜ Un−1,0 ˜ Un−1,0 ˜ Sn−1 Sn−1 Υ0 n−1,0 Υ1 n−1,0 f f cn−1 c˜n−1 hn−1,0 ˜1 Qn−1 1 Qn−1,0 Figure 3.2: Primitive case 1 contained in V and mod (V V ′) ε. As P : P − (V V ′) (V V ′) is a \ ≥ 0 0 \ → \ 1 covering of degree d, modulus of the annulus P − (V V ′) is ε/d which implies 0 \ that modulus of V ′ D ε/d. By Gr¨otzsch problem in [Ah66] (Section A in \ 1 ≥ Chapter III) we conclude that V ′ D contains a round annulus D D . \ 1 η(ε) \ 1 By considering the equipotentials of level η(ε) (obtained in the previous lemma) and the external rays landing at the dividing fixed points αn 1 andα ˜n 1 − − of the maps fcn−1 , and fc˜n−1 , we can form the favorite nest of puzzle pieces (3.1) introduced in Section 3.1.2. The hyperbolic distance between cn 1 andc ˜n 1 − − in the truncated primary wake containing cn 1 andc ˜n 1, W (η(ε)), is bounded − − by some M(ε) depending only on ε and the combinatorial class . That SL 43 is because cn 1 andc ˜n 1 belong to a finite number of truncated limbs which − − is a compact subset of this wake. Therefore, by Proposition 3.3, dilatation of the pseudo-conjugacy obtained in Theorem 3.2 is uniformly bounded by some constant K2(ε). χn qχn χn Let Qn,0 = Yn,0 and Pn,0 denote the last critical puzzle pieces obtained in the nest (3.1), and hn 1 = hn 1,0 denote the corresponding K2(ε)-q.c. pseudo- − − i χn i χn conjugacy obtained in Theorem 3.2. Components of fc−n−1 (Qn,0) and fc−n−1 (Pn,0) containing the little Julia sets Jn,i , for i =0, 1, 2,...,tn/tn 1 1, are denoted − − χn χn by Qn,i and Pn,i , respectively. Note that that tn/tn 1 is the period of the first − renormalization of fcn−1 . As the polynomials fcn−1 and fc˜n−1 also satisfy our combinatorial condition and a priori bounds assumption, there is a topological conjugacy, denoted by 1 ψn 1, between them which is obtained from extending Bc˜− −1 Bcn−1 onto the − n ◦ Julia set. χn χn Now we would like to adjust hn 1 : Qn Qn using dynamics of − → f tn/tn−1 : P χn Qχn , and f tn/tn− 1 : P χn Qχn . cn−1 n → n c˜n−1 n → n Let A0 denote the closure of the annulus Qχn P χn , and Ak , for k =0, 1, 2,... , n n \ n n ktn/tn−1 0 k denote the component of fc−n−1 (An) around Jn,0 by An. We can lift hn 1 by − tn/tn−1 χn tn/tn−1 χn 0 0 fc −1 : Q C and f : Q C to obtain a K -q.c. map g : A A n n → c˜n−1 n → 2 n → n 0 which is homotopic to ψn 1 relative boundaries of An. That is because by the − χn χn 0 external rays connecting ∂Pn to ∂Qn , the annulus An is partitioned into some topological disks and the two maps coincide on the boundaries of these topological disks. ktn/tn−1 k 0 ktn/tn−1 k 0 As fc −1 : A A and f : A A are holomorphic un- n n → n c˜n−1 n → n branched coverings, g can be lifted to a K -q.c. map from Ak to Ak , for every 2 n n k 1. All these lifts are the identity map in the B¨ottcher coordinate on the ≥ 44 boundaries of these annuli. Hence, they match together to K2-q.c. conjugate the two maps f tn/tn−1 : P χn J Qχn J , and f tn/tn−1 : P χn J˜ Qχn J˜ . cn−1 n \ n → n \ n c˜n−1 n \ n → n \ n Finally, we would like to extend this map further onto little Julia set Jn. This is a especial case of a more general argument presented below. Given a polynomial f with connected Julia set J, the rotation of angle θ on C J is defined as the rotation of angle θ in the B¨otcher coordinate on C J, \ \ 1 iθ that is, B− (e B ). By means of straightening, one can define rotations on c c the complement of the Julia set of a polynomial-like map. It is not canonical as it depends on the choice of straightening map. However, its effect on the landing points of external rays is canonical. Proposition 3.8. Let f : V V be a polynomial-like map with connected 2 → 1 Julia set J. If φ : V J V J is a homeomorphism which commutes with 1 \ → 1 \ f, then there exists a rotation of angle 2πj/(d 1), ρ , such that ρ φ extends − j j ◦ as the identity map onto J. Proof. Consider an external ray R landing at the non-dividing fixed point β0 of f. As this ray is invariant under f and φ commutes with f, we have f φ(R) = φ f(R) = φ(R). Thus, φ(R) is also invariant under f which ◦ ◦ implies that φ(R) lands at a non-dividing fixed point βj of f. Now, choose ρj such that ρ (φ(R)) lands at β . Let ψ denote ρ φ and R′ denote ρ (φ(R)). j 0 j ◦ j For such a rotation ρj, ψ also commutes with f and R′ is also invariant under f. The external ray R cuts the annulus V V into a quadrilateral I . The 1 \ 2 0,1 1 preimage f − (I0,1) produces d quadrilaterals denoted by I1,1,I1,2,...,I1,d, 45 n ordered clockwise starting with R. Similarly, the f -preimage of I0,1 produces n d quadrilaterals In,1,In,2,...,In,dn (also ordered clockwise starting with R). In the same way, the external ray R′ produces quadrilaterals denoted by In,j′ ordered clockwise starting with R′. First we will show that the Euclidean diameter of In,j (and In,j′ ) goes to zero as n tends to infinity. i Denote f − (V1) by Vi+1 and let di+1 denote the hyperbolic distance on the annulus V J. As I V , and the intersection of the nest of topological i+1 \ n,j ⊆ n+1 disks V is equal to J, the quadrilaterals I converge to the boundary of V J n n,j 1 \ as n goes to infinity. In order to show that the Euclidean diameter of these quadrilaterals go to zero, it is enough to show that their hyperbolic diameters in n 1 (V J, d ) stay bounded. Since f − :(V J, d ) (V J, d ) is an unbranched 1\ 1 n\ n → 1\ 1 n 1 n 1 covering of degree d − , therefore an isometry, and closure of f − (In,j) is a compact subset of V J, we conclude that I has bounded hyperbolic diameter 1\ n,j in (V J, d ). Finally, contraction of the inclusion map from (V J, d ) to n \ n n \ n (V J, d ) implies that I has bounded hyperbolic diameter in (V J, d ). 1 \ 1 n,j 1 \ 1 With the same type of argument, one can show that the hyperbolic distance between I and I′ inside (V J, d ) is also uniformly bounded. n,j n,j 1 \ 1 Since ψ is a conjugacy it sends In,j to In,j′ , and therefore, as w converges to J, w and ψ(w) belong to In,j and In,j′ with larger and larger index n. Combining with the above argument, we conclude that the Euclidean distance between these two points tends to zero. This implies that ψ can be extended as the identity map on the Julia set. A particular case of above proposition is that every homeomorphism which commutes with a quadratic polynomial-like map can be extended onto Julia set. 1 χn χn Applying the above lemma to ψn− 1 g with V1 = Qn , V2 = Pn , and an − ◦ 46 χn external ray connecting ∂Qn to Jn,0 we conclude that g can be extended as ψn 1 onto Jn,0. It also follows from proof of the above lemma that g and ψn 1 − − are homotopic relative J ∂Qχn . That is because the quadrilaterals obtained n ∪ n χn in the above lemma cut the puzzle piece Qn into infinite number of topological disks such that g and ψn 1 are equal on their boundaries. − χn Similarly, hn 1 can be adjusted on the other puzzle pieces Qn,i, and more- − χn χn over, hn 1 is homotopic to ψn 1 on Qn,i relative Jn,i ∂Qn,i. We will denote − − ∪ the map obtained from extending hn 1 onto little Julia sets Jn,i with the same − notation hn 1. − The final argument is to prepare hn 1 for the next step of the process. It − is stated in the following lemma. Lemma 3.9. The map hn 1 can be adjusted (through a homotopy) on a neigh- − borhood of iJn,i to a q.c. map hn′ 1 which maps Vn,i = Sn 1(Vn,i) onto ∪ − − ˜ Vn,i = Sn 1(Vn,i). Moreover, dilatation of hn′ 1 is uniformly bounded by a − − constant K (ε) depending only on ε. 2 Proof. The basic idea is move all of hn 1(Vn,i) simultaneously close enough to − little Julia sets Jn,i, and then move them back to Vn,i. We will do this more precisely below. Let Un,i denote Sn 1(Un,i) and Un,i denote Sn 1(Un,i). − − The annuli hn 1(Vn,i) Jn,i and Vn,i Jn,i have moduli bigger that ε/dK − \ \ where K is the dilatation of hn 1. Therefore, there exist topological disks − Ln,i with smooth boundaries and a constant r > 1 satisfying the following properties Ln,i hn 1(Vn,i) Vn,i • ⊂ − ∩ mod (L J ) r 1 • n,i \ n,i ≥ − mod (Vn,i Ln,i) ε/2dK and mod (hn 1(Vn,i) Ln,i) ε/2dK. • \ ≥ − \ ≥ 47 Now we claim that there exist q.c. maps χi : hn 1(Un,i), hn 1(Vn,i), Ln,i,Jn,i D5,D3,D2,D1 − − → with uniformly bounded dilatation. That is because all the annuli L J , n,i \ n,i hn 1(Vn,i) Ln,i, and hn 1(Un,i) hn 1(Vn,i) have moduli uniformly bounded − \ − \ − from below and above. The homotopy g : Dom(h) C, for t [0, 1], is t → ∈ defined as hn 1(z) if z / hn 1(Vn,i) − i − gt(z) := ∈ h 1 t ( χi n−1(z) 1)π χi− ( − sin | ◦ |− + 1) χi hn 1(z) if z hn 1(Vn,i). 3 4 ◦ − ∈ − It is straight to see that g0 hn 1, gt is a well defined homeomorphism for ≡ − every fixed t, and depends continuously on t fro every fixed z. For every 1 2 z ∂Vn,i, at time t = 1, we have g1(z) = χi− ( χi hn 1(z)) ∂Ln,i. That ∈ 3 ◦ − ∈ is, g maps ∂V to ∂L . For the returning part, we consider a q.c. map 1 n,i n,i Θ : U , V , L ,J D ,D ,D ,D , i n,i n,i n,i n,i → 5 3 2 1 and define gt+1 : Dom hn 1 C, for t [0, 1], as − → ∈ 1 g1(z) if z / g1− ( i Un,i) gt+1(z) := ∈ 1 t ( χi g1(z) 1)π 1 Θi− ( sin | ◦ |− + 1) Θi g1(z) if z g1− (Un,i ). √2 4 ◦ ∈ The homotopy gt for t [0, 2] is the desired adjustment and the map g2 : ∈ Dom hn 1 Range hn 1 is denoted by hn′ 1. − → − − Let ∆n 1,0 denote the Sn 1-preimage of the domain bounded by the equipo- − − tential of level η(ε) in the dynamic plane of fcn−1 . The multiply connected domain Ωn 1,0 is defined as − tn/tn−1 ∆n 1,0 Vn,itn−1 . − \ i=0 48 The domains ∆n 1,i and Ωn 1,i, for i = 1, 2,...,tn 1, are defined as the pull − − − i back of ∆n 1,0 and Ωn 1,i, respectively, under f − along the orbit of the critical − − point. Consider the map ˜ 1 hn 1,0 := Sn− 1 hn′ 1 Sn 1 : ∆n 1,0 ∆n 1,0, (3.4) − − ◦ − ◦ − − → − and then, ˜ i i hn 1,i := f − hn 1,0 f : ∆n 1,i ∆n 1,i. (3.5) − ◦ − ◦ − → − As these maps are compositions of two K1(ε)-q.c., a K 2(ε)-q.c., and possibly some holomorphic maps, they are q.c. with a uniform bound on their dilatation. By our adjustment in Lemma 3.9 we have hn 1,i maps Ωn 1,i onto Ωn 1,i. − − − Finally, the annulus Vn 1,0 Wn 1,0, with modulus bigger than ε/2 encloses − \ − Ωn 1,0 and is contained in Vn 1,0. This proves that the domain Ωn 1,0 is well − − − inside the disk Vn 1,0. Similarly, appropriate preimages of Vn 1,0 Wn 1,0 un- − − \ − i der the conformal maps f − introduce definite annuli around Ωn 1,i which are − contained in Vn 1,i. In this case, the topological disks Dn,i are the domains Vn,i − which contain the little Julia sets Jn,i well inside themselves. Case B: Here, fcn−1 is satellite renormalizable and its second renormalization is of primi- tive type. Let αn 1 denote the dividing fixed point of fcn−1 , and αn the dividing − fixed point of it’s first renormalization f − . In this situation, the Julia set R cn 1 of the primitive renormalization f − , and its forward images under f − can R cn 1 cn 1 get arbitrarily close to αn 1 (which is a non-dividing fixed point of fcn−1 ). − R Our idea is to skip the satellite renormalization in this case which essentially imposes the secondary limbs condition on us. 49 Consider an equipotential of level η(ε) contained in Sn 1(Wn 1,0), the ex- − − ternal rays landing at αn 1, and the external rays landing at the fcn−1 -orbit − η(ε) of αn (see Figure 3.3). These rays and the equipotential E depend holo- morphically on the parameter within the secondary wake W (η) containing the tn/tn−1 parameter cn 1. Let us denote fcn−1 by g for simplicity of notation through − this case. Ray landing at αn 1 − χ1 Qn 0 A1 0 C0 fcn−1 b 0 Aj 0 Bi Ray landing at αn Figure 3.3: Figure of an infinitely renormalizable Julia set. The first renormal- ization is of satellite type and the second one is of primitive type. The puzzle χ1 piece at the center, Qn , is the first puzzle piece in the favorite nest. Now, using the above rays and equipotential, we introduce some new puzzle 0 pieces. Let Y0 , as before, denote the puzzle piece containing the critical point and bounded by E(η), the external rays landing at αn 1 and the external rays − landing at fcn−1 -preimages of αn 1 (i.e. at all ωαn 1 with ω a dth root of unity). − − 0 The External rays landing at αn and their g-preimages, cut the puzzle piece Y0 into finitely many pieces. Let us denote the one containing the critical point 50 0 0 by C0 , the non-critical ones which have a boundary ray landing at αn by Bi 0 and the rest of them by Aj (these ones have a boundary external ray landing at some ωαn). 0 The g-preimage of Y0 along the postcritical set is contained in itself. As all processes of making modified principal nest and the pseudo-conjugacy in Theorem 3.2 are based on pullback arguments, the same ideas are applicable here. The only difference is that we do not have equipotentials for the second renormalization. As we will see in a moment, certain external rays and part 0 of the equipotential bounding Y0 will play the role of an equipotential for the second renormalization of fcn−1 . By definition of satellite and primitive renormalizability, every gn(0) be- longs to Y 0, for n 0, and there is a first moment t with gt(0) A0 (by 0 ≥ ∈ 1 0 t rearranging the indices if required). Pulling back A1 under g along the critical orbit, we obtain a puzzle piece Qχ1 0, such that C0 Qχ1 is a non-degenerate n ∋ 0 \ n 0 annulus. That is because C0 is bounded by the external rays landing at αn χ1 0 and their g-preimage. Therefore, if Qn intersects ∂C0 at some point on the rays, orbit of this intersection under gk, for k 1, stays on the rays landing at ≥ χ1 0 αn. This implies that image of Qn can not be A1. Also, they can not intersect at equipotentials as they have different levels. Now, consider the first moment m χ1 χ1 m m > t when g (0) returns back to Qn , and pullback Qn under g along the χ1 m critical orbit to obtain Pn . The map g is a unicritical degree d branched χ1 χ1 covering from Pn onto Qn . This introduces the first two pieces in the favorite nest. The rest of the process to form the whole favorite nest is the same as in Section 3.1.2. tn/tn−1 0 tn/tn−1 0 Consider the map fc −1 : Y fc −1 (Y ), and the corresponding tilde n 0 → n 0 one. One applies Theorem 3.2 to these maps, using the favorite nests intro- 51 duced in the above paragraph, to obtain a q.c. pseudo-conjugacy tn/tn−1 0 tn/tn−1 0 hn 1 : fc −1 (Y0 ) fc −1 (Y0 ). − n → n The equipotential of level η(ε), the external rays landing at αn 1, and the − external rays landing at the fcn−1 -orbit of αn depend holomorphically on the parameter within the secondary wake W (η) containing the parameter cn 1. − Therefore, by Proposition 3.3, the dilatation of hn 1 depends on the hyperbolic − distance between cn 1 andc ˜n 1 within one of the finite secondary wakes W (η) − − under our combinatorial assumption. Thus, it only depends on the a priori bounds ε and the combinatorial condition. tn/tn−1 j tn/tn−1 j − 0 0 As fc −1 : fcn−1 (Y0 ) fcn−1 (Y0 ), for j = 1, 2,...,tn/tn 1 1, is n → − − univalent, we can lift hn 1 on other puzzle pieces as − j j tn/tn−1 j 0 tn/tn−1 j 0 hn 1 := fc−−1 hn 1 fc −1 : fc −1 − (Y0 ) fc −1 − (Y0 ) − n ◦ − ◦ n n → n for these j’s. Since all these maps match the B¨ottcher coordinates, they fit together to build a q.c. map from a neighborhood of J(fcn−1 ) to a neighborhood of J(fcn−1 ). Further, it can be extended as the identity map in the B¨ottcher coordinates to a q.c. map from the domain bounded by equipotential Eη(ε) to the corresponding tilde domain. Finally, by Lemma 3.15, we adjust hn 1 to obtain a q.c. map hn′ 1,0 that − − satisfies ˜ hn′ 1(Sn 1(Vn+1,itn−1 )) = Sn 1(Vn,itn−1 ), for i =0, 1, 2,...,tn+1/tn 1 1. − − − − − Now, ∆n 1,0 is defined as Sn 1-pullback of the domain bounded by the − − η(ε) equipotential E . The domain Ωn 1,0 is − tn+1/tn−1 1 − ∆n 1,0 Vn+1,itn−1 . − \ i=0 52 The regions ∆n 1,i and Ωn 1,i, for i =1, 2, 3,...,tn 1, are pullbacks of ∆n 1,0 − − − − i and Ωn 1,0 under f , respectively. Like in the previous case, hn 1,i is defined − − as in Equations (3.4) and (3.5) and satisfies hn 1,i(Ωn 1,i)= Ωn 1,i, for i =1, 2,...,tn 1. − − − − For the same reason as in Case A , Ωn 1,i is well inside the disk Dn 1,i := Vn 1,i. − − − Case C : Here, fcn−1 is twice satellite renormalizable. The argument in this case relies more on the compactness of the parameters under consideration rather than a dynamical discussion. Figure 3.4: A twice satellite renormalizable Julia set drawn in grey. The dark part is the Julia Bouquet B2,0. 53 Little Julia sets J1,i of the first renormalization of fcn−1 touch at the dividing fixed point αn 1 of fcn−1 . Note that αn 1 is one of the non-dividing fixed − − points of the first renormalization of fcn−1 . The union of these little Julia sets is called Julia bouquet and denoted by B1,0. Similarly, the little Julia sets J2,i, for i = 0, 1, 2,..., (tn+1/tn 1) 1, of the second renormalization of fcn−1 − − are organized in pairwise disjoint bouquets, B2,j, touching at a periodic point. That is, each B2,j consists of tn+1/tn little Julia sets J2,i touching at one of their non-dividing fixed points. As usual, B2,0 denotes the bouquet containing the critical point. See Figure 3.4. By an equipotential of level η(ε) (which encloses Sn 1(Wn 1,0)) and the − − external rays landing at αn 1, we form the puzzle pieces of level zero. Recall − 0 that Y0 denotes the one containing the critical point. The following lemma states that the bouquets are well apart form each other. Lemma 3.10. For parameters in a finite number of truncated secondary limbs, modulus of the annulus Y 0 B is uniformly bounded above and below. 0 \ 2,0 To prove this lemma we first need to recall some definitions. Let X and Y be compact subsets of C equipped with the Euclidean metric d. The Hausdorff distance between X and Y is defined as dH (X,Y ) := inf ε : Y Bε(X), and X Bε(Y ) . ε { ⊂ ⊂ } The space of all compact subsets of C endowed with this metric is a complete metric space. A set valued map c X , with X compact in C, is upper semi-continuous → c c if c c implies that X contains the Hausdorff limit of X . n → c cn A family of simply connected domains Uλ depends continuously on λ if there exit choices of uniformizations ψ : D U continuous in both variables. λ 1 → λ 54 A family of polynomial-like maps (P : V U ,λ Λ), parametrized on λ λ → λ ∈ a topological disk Λ, depends continuously on λ if Uλ is a continuous family of simply connected domains in C, and for every fixed z C, P (z) depends ∈ λ continuously on λ where ever it is defined. Proposition 3.11. Let (P : V U ,λ Λ) be a continuous family of λ λ → λ ∈ polynomial like maps with connected filled Julia sets K . The map λ K is λ → λ upper semi-continuous. Proof. Assume that λ λ, K := K(P ), and K := K(P ). To prove that n → λn λn λ λ K := lim Kλn is contained in Kλ, it is enough to show that for every ε > 0, K B (K ) for sufficiently large n. λn ⊆ ε λ To see this, assume that z / B (K ). If z / V , then by continuous ∈ ε λ ∈ λ dependence of V on λ, z / K for large n. If z V , then there exists a λ ∈ λn ∈ λ positive integer l with P l (z) U V . As P : V U converges to λ ∈ λ \ λ λn λn → λn P : V U , P l (z) or P l+1(z) belongs to U V . Therefore, z / K . λ λ → λ n n λn \ λn ∈ λn Proof of Lemma 3.10. Let cn 1 be a twice satellite renormalizable parameter − in a truncated secondary limb. Consider the external rays landing at the 0 dividing fixed point α of f − and it’s preimages under f − . Let X n R cn 1 R cn 1 0 denote the puzzle piece obtained from these rays which contains the critical 2 0 point. As f − is twice satellite renormalizable, f − : X C is a branched cn 1 R cn 1 0 → 0 covering map over its image. One can consider a continuous thickening of X0 , described in Section 3.1.2, to form a continuous family of polynomial-like maps parametrized over this truncated limb. The little Julia set of this map was denoted by J2,0 and the Julia bouquet B2,0 is the connected component of tn+1/tn−1 1 − fcn−1 (J2,0) i=0 containing the critical point. 55 For cn 1 in a finite number of truncated secondary limbs, the Julia bouquet − B2,0 is unoin of a finite number of little Julia sets. Hence, by above lemma, it depends upper semi-continuously on cn 1. As B2,0 is contained well inside the − 0 interior of Y0 for such cn 1 which belongs to a compact set of parameters, we − conclude that modulus of Y 0 B is uniformly bounded below. 0 \ 2,j To see that this modulus is uniformly bounded above, one only needs to observe that αn and 0 belong to B2,0 and are strictly apart from each other for these parameters. By Lemma 3.10 there are simply connected domains Ln′ 1 Ln 1 and − ⊆ − Ln′ 1 Ln 1 such that moduli of the annuli − ⊆ −